Abstract: The mitochondrial genome (mitogenome) is one of the most widely used markers for phylogenetic analysis. Compared with whole-genome data, mitogenome data are less expensive to obtain and easier to manipulate. However, compositional bias and accelerated evolutionary rate reduce the effectiveness of the mitogenome in determining insect phylogeny. This study shows that mitogenome data are not suitable to reconstruct deep holometabolan evolution, even with a most comprehensive data coding scheme and the more realistic CAT model. For the deep levels of divergence within Holometabola, protein-coding genes only retain weak phylogenetic signals, leading to peculiar interordinal relationships. Consensus relationships in the Holometabola phylogeny, such as the monophyly of Holometabola, the most basal position of Hymenoptera, and the sister group relationship between the Strepsiptera and Coleoptera were rarely resolved in our analyses. The relationships of the holometabolan groups as inferred by mitogenomes are highly vulnerable to gene types, data coding regimes, model choice, and optimality criteria, and no consistent alternative hypothesis of Holometabola's relationships is supported. Thus, we suggest that the slowly evolving nuclear genes or genome-scale approaches may be better options for resolving deep-level phylogeny of Holometabola.

Key words: data coding, Holometabola, mitochondrial genome, phylogeny